Simulation In The System Engineering Process “System Dynamics for System Engineering Process” Dr. Warren W. Tignor Ph.D. SAIC Second Annual Workshop on Web Enabled Modeling and Simulation Arlington, VA October 12-15, 2004 Agenda • WebSim 2003 Continuity to WebSim 2004 • System Dynamics (SD) Introduction • SD and UML Synergy • Case Study • Findings • Conclusions • Literature Survey WebSim 2003 Continuity to WebSim 2004 WebSim 2003 Keynote Speaker Items Shaping Evolution of the Global Information Grid (GIG) Dawn Meyerriecks Chief Technology Officer Defense Information Systems Agency [email protected] Design Imperatives the New Imperatives How to Improve Design of large, complex Net- centric capabilities? How to Lower Risks inherent in designing and Dynamic deploying large, complex Net-centric capabilities? • In early stages of design: Simulated – Dynamic models could be used to evaluate alternative concepts Services • In later stages of deployment: – Simulated services and applications could be used for fault ID and isolation of operational problems UNCLASSIFIED New Approach the New M&S Approach Dynamic Models • Dynamic models would: – Simulate actions taking place within the GIG (Allow “what if” exploration) – Evolve as greater detail is added – Provide Independent Validation and Verification (IV&V) Bottom Line: Ensure GIG capabilities function as envisioned UNCLASSIFIED Summary of WebSim 2003 Issues • Help improve design and lower risk at – Early in design – Later in implementation • Support “What ifs” • Evolve as add details • Support IV&V • Show continuous dynamics of the system System Dynamics (SD) Introduction System Engineering & System Dynamics (Modeling & Simulation) Systems Engineering • Organized interdisciplinary approach to enable systems. • Focused on complete development cycle from requirements, to design and system validation. • Integrated team effort that proceeds from concept to production, to operation. • Considers both the business and the technical needs of all customers. Modeling & Simulation • Modeling investigates the important characteristics of a system in the early phases of the development cycle as well as in the operational phase. • Simulation identifies bottlenecks in a system, evaluate system capacity, utilization and cost, and generate operational plans and schedules. UML Modeling Diagrams • Structural Diagrams -Class, Object, Component, and Deployment • Behavior Diagrams - Use Case, Sequence, Activity, Collaboration, and State chart • Model Management Diagrams - Packages, Subsystems, and Models. SD Overview • System Dynamics – a computer based approach that: – Models complex physical & social systems – Experiments with design policies for improved performance – Abstracts the system around a problematic pattern – Uses feedback loops as • Building blocks • Means to explain its behavior. System Dynamics (SD) Perspective • Holistic - System dynamicists look at things as a whole. • Non-reductionist - Unlike others, who study the world by breaking it up into smaller and smaller pieces. SD Scope • Understand system’s basic structure • Understand system’s behavior • Use computer model to: – Perform complex simultaneous calculations – Compare results to human mental model The SD Methodology Steps • Identifies a problem • Develops a dynamic hypothesis-(behavior) • Builds a computer simulation model- (structure & behavior) • Tests the model (reference to real-world) • Devises and tests alternative policies that alleviate the problem-(what-ifs) Some SD Domains • Corporate planning and policy design • Public management and policy • Biological and medical modeling • Energy and the environment • Theory development in the natural and social sciences • Dynamic decision making • Complex nonlinear dynamics, emergence SD and UML Synergy SD & UML Synergy • Unified Modeling Language™ specification addresses need to model the structure and performance of information systems (IS). • Preliminary studies show UML2™ may not close performance simulation gap . • System Dynamics (SD) is based on system structure and behavior with simulation capability. Framing the Case Study Hypothesis, Research Method, & Case Hypothesis • SD will leverage UML structure and behavior models • System Engineers (SEs) may apply SD to IS structure and behavior simulation • SD is a candidate to close gap in UML™ simulation capability Research Method and Design • Used a case study of a real-life problem described in UML and a non-system dynamics simulation language, Simulation Modeling Language (SimML), and recasts the UML into a System Dynamics model using Vensim. The Case • Arief (2001) documents a British Telecom (BT) Intelligent Network (IN) case to: – Size the server capacity needed to make and receive calls – Accommodate capability for caller id, call blocking, billing and more. UML™ to Simulation Path – Arief (2001) Simulation from UML™ - Arief (2001) SimML Parser generates uses Simulation UML Tool uses Program generates Textual notation Architecture of BT IN – Arief (2001) (WAN, 34 Mbps, 50 ms latency) (60-1000, geo. Distributed) (approx. 10) (LAN, 100 Mbps, 1ms latency) (3 x 10E3 to 3 x 10E6 messages/sec) (callers 10E5 to 10E8) UML™ Class Diagram of BT IN – Arief (2001) {bandwidth ~unknown and latency ~0ms} UML™ Activity Diagram – Arief (2001) UML™ Sequence Diagram– Arief (2001) T i m e Variables Used – Arief (2001) Name Type Explanation Approx. Values interArr Exponential The inter arrival time of the calls 1000 calls/sec lookupTime Exponential The time taken by the Name Server to Dependent Variable lookup for call identities ~ .0002 ms avg readTime Exponential The time taken to perform the barOutgoing flag evaluation .00023 ms avg searchTime Exponential The time taken to check the blacklist .0057 ms avg localDelay Exponential The network latency for local call objects ~ 0 lanDelay Exponential The network latency for LAN call objects 1 ms wanDelay Exponential The network latency for WAN call objects 50 ms rndCallGen Uniform Used for randomly generating the local/LAN/WAN call types Run time integer Duration of simulation 100,000 ms SimML results – Arief (2001) lookup average times all calls Time local LAN WAN done time avgTime 20.00 523.80 526.02 719.90 299603 1.77E8 589.20 19.99 499.13 501.38 695.36 299748 1.69E8 564.58 19.98 474.39 476.76 670.77 299889 1.62E8 539.93 19.97 449.67 452.12 646.22 300053 1.55E8 515.29 19.96 424.88 427.49 621.68 300223 1.47E8 490.65 19.95 400.05 402.79 597.00 300353 1.40E8 465.91 19.94 375.25 378.10 572.33 300511 1.33E8 441.19 19.93 350.44 353.33 547.61 300650 1.25E8 416.42 19.92 325.60 328.58 522.87 300804 1.18E8 391.64 19.91 300.71 303.76 498.13 300945 1.10E8 366.83 19.90 275.83 278.94 473.33 301091 1.03E8 341.99 SD Stock and Flow Diagram Error Calls Calls at Calls UID'd Calls Made Switch connect assign UID bar call blacklist call Calls Calls Received Connected ring call hangup Calls Not Answered end ring call SD Stock and Flow Diagram with Parameters % <TIME STEP> Calls <Time> fraction calls unbarred fraction calls unblacklisted Switch Calls ID'd Calls Dialed Calls Ringed Messages connect lookup unbarred unblacklisted calls calls avg lookup avg arrival time rate avg read time avg avg # sites avg LAN WAN avg search time network latency network latency processing rate avg LOCAL network latency SD Model Percentage Performance with various average Lookups % 98.89 93.89 88.89 500 1925 3350 4775 6200 7625 9050 Time (ms) "%" : 1call per ms & 2E-05 avg lookup Dmnl "%" : 1call per ms & 2E-03 avg lookup Dmnl "%" : 1call per ms & 2E-04 avg lookup Dmnl SD Model Switch Messages accumulations with various average Lookups Switch Messages 20.00 10.00 0.0100 500 1925 3350 4775 6200 7625 9050 Time (ms) Switch Messages : 1call per ms & 2E-05 avg lookup calls Switch Messages : 1call per ms & 2E-03 avg lookup calls Switch Messages : 1call per ms & 2E-04 avg lookup calls SD Model Calls ID’d accumulations with various average Lookups Calls ID'd 3.300 2.206 1.112 500 1925 3350 4775 6200 7625 9050 Time (ms) Calls ID'd : 1call per ms & 2E-05 avg lookup calls Calls ID'd : 1call per ms & 2E-03 avg lookup calls Calls ID'd : 1call per ms & 2E-04 avg lookup calls SD Model Calls Dialed accumulations with various average Lookups Calls Dialed 107.68 80.55 53.43 500 1925 3350 4775 6200 7625 9050 Time (ms) Calls Dialed : 1call per ms & 2E-05 avg lookup calls Calls Dialed : 1call per ms & 2E-03 avg lookup calls Calls Dialed : 1call per ms & 2E-04 avg lookup calls Model Calls Ringed accumulations with various average Lookups Calls Ringed 9,889 5,166 444.45 500 1925 3350 4775 6200 7625 9050 Time (ms) Calls Ringed : 1call per ms & 2E-05 avg lookup calls Calls Ringed : 1call per ms & 2E-03 avg lookup calls Calls Ringed : 1call per ms & 2E-04 avg lookup calls SD Model Switch Messages accumulations with various average Lookups & 3calls/ms Switch Messages 20.00 10.00 0.0100 500 1925 3350 4775 6200 7625 9050 Time (ms) Switch Messages : 3call per ms & 2E-04 avg lookup calls Switch Messages : 1call per ms & 2E-05 avg lookup calls Switch Messages : 1call per ms & 2E-03 avg lookup calls Switch Messages : 1call per ms & 2E-04 avg lookup calls Major Findings & Conclusion Major Findings and Significance • UML and SD Models – SD VENSIM models do not solve translation of static UML diagrams into executable language as SimML automated this step – SD has power to support this concept – tools moving to XML interface Major Findings and Significance • Integrated View of Structure and Behavior – SD VENSIM shows integrated view of the structure and behavior of BT IN model • Lacking in UML™ models & • Lacking in SimML language – SD VENSIM shows interdependence of structure and behavior concurrently • Using stock and flow diagrams & • Using simulation results. Major Findings and Significance • Quality of Service (QoS) Requirements – SD models address QoS requirements such as availability or maintainability as auxiliary parameters either in soft (e.g. High medium, low) or hard form (e.g., .15 units).